Implementation of Anatomical Navigators
for Real Time Motion Correction in
Diffusion Tensor Imaging
Alkathafi A. Alhamud
Department of Human Biology
University of Cape Town
Thesis presented for the degree of
Doctor of Philosophy
February 2012
Abstract
Thesis: Implementation of Anatomical Navigators for Real Time Motion Correction in
Diffusion Tensor Imaging
Author‟s Name: Alkathafi A. Alhamud Date: 10-02-2012
Diffusion Tensor Imaging (DTI) is used to detect microstructural changes in the human brain.
Although echo planar imaging (EPI) has been implemented in diffusion pulse sequences to
minimize the effect of subject motion, motion artifacts may reappear in the diffusion volume
images when scanning the brain repeatedly with different diffusion directions and strengths.
Both retrospective and prospective motion correction approaches have been implemented in DTI
to correct for misalignments in diffusion volume images. Most of these methods rely on the
contrast differences between the diffusion weighted images and a target image, which renders the
registration for motion correction inaccurate, especially in acquisitions with high diffusion
weightings. In order to develop a motion correction method that is independent of the diffusion
contrast, a separate motion tracking technique has been introduced using a volumetric 3D-EPI
navigator. This technique performs prospective motion correction in diffusion weighted images
without having to reacquire volumes during which motion occurred, unless motion exceeded
some pre-set thresholds. The additional scan time for the navigator and feedback is only 526 ms
per diffusion volume, which takes 9500 ms to acquire.
The navigated diffusion sequence (vNav) was validated using a water phantom, and in vivo in
sixteen children (aged 5-6 years) and 6 adults. A multiecho MPRAGE sequence was also
acquired in paediatric and adult subjects. FreeSurfer was used to automatically extract volumes
of interest (VOI‟s). The mean and the histogram-derived measurements of the FA and MD for
the whole brain (WBH) and for different VOI‟s were analysed. Our results show that adding the
navigator does not alter the DTI data. In the paediatric and adult scans acquired using the
standard diffusion sequence, subject head motion caused significant changes in all the DTI
measures, except FA of the whole brain white matter in children. Retrospective motion
correction did not recover diffusion measures but instead resulted in decreased FA and generated
spurious fiber tracts in the corpus callosum. The DTI measures are recovered substantially in the
prospective motion corrected data acquired using the navigated sequence.
ii
Contents
Contents .......................................................................................................................................... ii
List of Figures ................................................................................................................................. v
List of Tables ............................................................................................................................... xiii
Preface........................................................................................................................................... xv
Acknowledgments....................................................................................................................... xvii
1 Introduction ............................................................................................................................. 1
1.1 Background and Motivation ............................................................................................. 2
1.2 Theory .............................................................................................................................. 3
1.2.1 MRI Physics ............................................................................................................ 3
1.2.2 Pulse Sequences ...................................................................................................... 5
1.2.3 Diffusion ............................................................................................................... 11
1.2.4 Diffusion Weighted Imaging (DWI) Pulse Sequence ........................................... 12
1.2.5 Elements of a DWI sequence ................................................................................ 16
1.3 Diffusion Tensor Imaging (DTI) .................................................................................... 17
1.3.1 From DWI to DTI ................................................................................................. 17
1.3.2 Effects of Subject Head Motion in DTI ................................................................ 18
1.3.3 Motion Correction in conventional MRI .............................................................. 19
1.3.4 Navigator Techniques in Diffusion MRI .............................................................. 22
1.3.5 Motion Correction in DTI ..................................................................................... 23
2 Volumetric Navigators for Real Time Motion Correction in Diffusion Tensor
Imaging………… ................................................................................................................. 27
2.1 Introduction .................................................................................................................... 28
2.2 Materials and Methods ................................................................................................... 31
2.2.1 Preparation of the 3D-EPI Navigator Sequence ................................................... 31
iii
2.2.2 Diffusion Pulse Sequence with 3D-EPI Navigator ............................................... 31
2.2.3 Reacquisition......................................................................................................... 33
2.2.4 Feedback and Combination of Motion Parameters ............................................... 34
2.2.5 Experimental Protocol .......................................................................................... 34
2.2.6 Diffusion Data Processing .................................................................................... 36
2.3 Results ............................................................................................................................ 37
2.3.1 The Influence of the Navigator on the Diffusion Sequence ................................. 37
2.3.2 Subject Motion, Motion Correction and Reacquisition ........................................ 39
2.4 Discussion ...................................................................................................................... 47
2.5 Conclusion ...................................................................................................................... 51
2.6 Acknowledgements ........................................................................................................ 52
Addendum to Chapter 2 ................................................................................................................ 53
A.1 Rotation of the Diffusion Table ..................................................................................... 54
A.2 Eddy current correction .................................................................................................. 55
A.3 Details of Patient Motion ............................................................................................... 61
3 Potential Misinterpretation of Abnormal Diffusion Tensor Imaging Derived Metrics in the
Presence of Motion ............................................................................................................... 63
3.1 Introduction .................................................................................................................... 65
3.2 Material and Methods: ................................................................................................... 67
3.2.1 Navigated diffusion pulse sequence with 3D-EPI Navigator and reacquisition ... 67
3.2.2 MRI Data Acquisition ........................................................................................... 67
3.2.3 Data Processing ..................................................................................................... 68
3.3 Results ............................................................................................................................ 71
3.4 Discussion ...................................................................................................................... 84
3.5 Conclusions .................................................................................................................... 88
3.6 Acknowledgements ........................................................................................................ 88
iv
4 Application of the EPI-Navigated Diffusion Tensor Imaging Sequence to Paediatric
scans……….. ........................................................................................................................ 89
4.1 Introduction .................................................................................................................... 91
4.2 Materials and Methods ................................................................................................... 92
4.2.1 Navigated diffusion pulse sequence with 3D-EPI Navigator and reacquisition ... 92
4.2.2 MRI Data Acquisition ........................................................................................... 92
4.2.3 Data Processing ..................................................................................................... 94
4.3 Results ............................................................................................................................ 95
4.4 Discussion .................................................................................................................... 106
4.5 Conclusion .................................................................................................................... 110
4.6 Acknowledgements ...................................................................................................... 111
5 Discussion ........................................................................................................................... 112
5.1 Comparison of the vNav to Current Techniques .......................................................... 112
5.2 Effect of Motion on the B-Matrix ................................................................................ 116
5.3 DTI Data Analysis with the Standard and the Navigated Diffusion Sequences .......... 118
5.3.1 Validation of the Navigator ................................................................................ 118
5.3.2 Age-related Differences in Whole Brain Histogram FA .................................... 118
5.3.3 Effects of Motion and motion correction on Whole Brain Histogram FA in Adults
and Children ........................................................................................................................ 119
5.3.4 Effects of Motion and motion correction on Regional FA and MD in Adults and
Children……………………………………………………………………………………119
5.4 Limitations to this Study .............................................................................................. 121
6 Conclusion .......................................................................................................................... 123
References ................................................................................................................................... 126
v
List of Figures
1.1 The basic elements of the gradient echo MR imaging sequence. Amplitude is shown
vertically, time horizontally. RF is the excitation pulse, Gss is the slice selective gradient,
GPE is the phase encoding gradient, GFE is the frequency encoding gradient. The sequence
is repeated for many different values of GPE………………………………….…..………6
1.2 Spin-echo sequence with the 180o refocused RF pulse. Following the RF pulse, the spins
are initially in phase (b) but dephase during the period labeled (c). At time (d), a 180o
pulse is applied. During period (e) the spins recover their phase. The signal is maximal at
(f) whereafter the spins again diphase…………………………...………………………..7
1.3 The spin-echo pulse sequence with the crusher gradients on both sides of the 180o
refocusing RF pulse. The crusher gradients are used to dephase the unwanted
magnetization (FID signal) created by the refocusing pulse…………………...…………8
1.4 (a) In a „Blipped‟ single-shot GE-EPI sequence a small phase-encoding gradient „blip‟ is
placed at each readout gradient reversal; (b) In a „Unblipped‟ single-shot GE-EPI a
constant phase-encode is applied continuously along the readout gradient reversal. In
both modalities the whole image is acquired following a single
excitation.………………………………………………………...………………….……9
1.5 CHESS (CHEmical Shift Selective) sequence diagram. The x axis is the time in µs, the y
axis is the magnitude. From top to bottom, the crusher gradients are applied in all three
directions, X,Y and Z, respectively. The bottom trace illustrates the timing of the fat
selective (CHESS) RF pulse. The crushers are used to destroy the phase coherence of
transverse fat magnetization. This diagram was generated by the Sequence Development
Environment Software (IDEA) provided by Siemens by simulating the DTI pulse
sequence……………………………………………………………………………...…..11
1.6 Illustration of the main components of the Stejskal and Tanner spin echo diffusion
sequence that is designed to measure the diffusion coefficient along the direction of the
diffusion gradient. δ is the duration of the diffusion gradient, Δ is the center-to-center
vi
spacing, and G is the magnitude of the diffusion gradient in mT/m. Other sequence
elements were omitted for clarity………………………………………………………..13
1.7 Typical motion artefact: ghost and signal variation across diffusion images (Le Bihan et
al., 2006)……………………………………………..…………………………………..14
1.8 Images showing geometrical distortions resulting from eddy currents that are associated
with the use of strong diffusion gradients: contraction (top right), shift (bottom left), and
shear in the bottom right (Le Bihan et al., 2006)…………………………….….……….15
1.9 Diagram of the standard twice-refocused spin echo (TRSE) sequence that successfully
reduces the effect of eddy currents (Reese et al., 2003). Shown are the 90o RF pulse for
excitation and the two 180o refocusing RF pulses. G is the strength of the diffusion
gradients. The duration of the diffusion gradients are δ1, δ2, δ3, and δ4, respectively, and
the data is acquired using an EPI readout. Other sequence elements were omitted for
clarity………………………………………………………………………………….....15
1.10 Illustration of the sequence elements of the single-shot DTI pulse sequence used in the
current study. The elements are (a) the CHESS pulse, (b) additional readout for phase
correction, (c) the crusher gradients, (d) two 180o refocusing pulses, (e) diffusion
gradients implemented using the twice-refocused pulse technique. The elements of the
blipped EPI readout are indicated by f. The x axis represents the time and the y axis
represents the magnitude of the different components. This diagram was generated by the
Sequence Development Environment Software (IDEA) provided by Siemens by
simulating the DTI pulse sequence.…………………………………………………...…16
1.11 Data affected by signal dropouts are indicated by circles. Information for these slices is
typically lost, leading to reduced SNR or biased results. The DTI acquisition for this
example consisted of 34 measurements or diffusion volumes, each with 72 slices, with a
maximum b value of 1000 sec mm-2
. The current volume was # 12……..…………...…19
1.12 Comparison of the sequence timing diagram of the standard diffusion sequence without
navigators (a) and the diffusion sequence with navigators inserted between the diffusion
volumes (b). The diffusion protocol for this example was: 64 slices for each volume, one
dummy scan, 7 diffusion volumes, one b0 volume image, 6 diffusion gradient directions
vii
with b value 1000 sec mm-2
. The scan time with the standard sequence was ~ 68 sec and
with the navigated sequence was ~ 72 sec. The navigator acquisition matrix was 32 × 32
× 28 and the acquisition time for each navigator was 406 ms. The x axis represents the
time and the y axis represents the magnitude of the sequence elements. This diagram was
generated using the Sequence Development Environment Software (IDEA) provided by
Siemens by simulating the DTI pulse sequence………...……………………………….26
2.1 Flowchart of the modified diffusion pulse sequence with interleaved 3D-EPI navigator.
Data is transferred to ICE, registration is performed by PACE, and the position and
gradient system is updated before acquisition of the next diffusion volume…………….32
2.2 (a) The MD map for slice 38 of a stationary water phantom scanned using the basic
diffusion sequence, and the difference between the MD map of this slice and the MD map
of the same slice for scan (b) W_basic2, (c) W_basic3, (d) W_vNav1, (e) W_vNav2, and
(f) W_vNav3, where W denotes water phantom scans, basic denotes scans acquired using
the basic diffusion sequence, and vNav denotes scans acquired using the navigated
prospective motion corrected diffusion sequence. All color bars have units 10−3
mm2
s−1
………………………………………………………………………………..……….38
2.3 Histograms of the averaged MD for the three scans over the whole volume of the water
phantom for the navigated prospective motion corrected diffusion sequence (vNav) and
the basic diffusion sequence (basic)……………………………………………….…….38
2.4 Normalized whole brain histograms (WBHs) of FA for two subjects (2 and 5) for the at
rest (NoMo) scans acquired with the basic diffusion sequence and with the navigated
sequence without prospective motion correction (vNav_NoCo). The plots in the bottom
row are the corresponding motion parameters that were estimated in ICE by PACE
during the NoMo_vNav_NoCo scans……………………………………………………39
2.5 Comparison of motion parameter estimates generated by PACE and retrospective (retro)
motion correction for the first subject. a and d show the motion parameters that were
estimated in ICE by PACE for the Mo_vNav_NoCo and Mo_vNav_all scans,
respectively. b and e show the retrospective motion estimates for the Mo_vNav_NoCo
scan using SPM and FLIRT, respectively, and same for the basic diffusion sequence in c
and f. Mo denotes a scan with motion, vNav the navigated diffusion sequence, NoCo
viii
without prospective motion correction, and all a scan with prospective motion correction
and reacquisition enabled. Subjects moved upon verbal instruction, five to six times
during the scan……………………………………………………………………..…….40
2.6 FA and MD maps of slice 80 for the first subject for five different acquisitions, as well as
results of retrospective motion correction in the scan acquired using the basic sequence
using SPM (Mo_basic_retro_SPM) and FLIRT (Mo_basic_retro_FLIRT), respectively.
Data acquired in the Mo_vNav_all scan have been analyzed both without
(Mo_vNav_noReAq) and with (Mo_vNav_ReAq) reacquisition. The two yellow circles
on the FA maps demonstrate reduced blurring in the scan with reacquisition compared to
the scan without reacquisition. All the maps are coregistered to the T1 space…………..42
2.7 Comparison of the normalized WBH-FA (a-d) and WBH-MD (e-h) for the first subject
for different scans: (a,e) Comparison of at rest scans acquired using both the basic and
navigated sequence to scans with motion and no prospective motion correction acquired
using both the basic and navigated sequence; (b,f) Effect of retrospective motion
correction using SPM on the scans acquired without prospective motion correction; (c,g)
Effect of retrospective motion correction using FLIRT on the scans acquired without
prospective motion correction; and (d,h) prospective motion corrected scans acquired
using the navigated sequence, both without and with reacquisition……………………..43
2.8 Comparison of the normalized WBH-FA in three subjects for an at rest baseline scan
(NoMo_basic) compared to a scan with motion and retrospective motion correction
where the suffices BE and AE, respectively, denote before and after elimination of
corrupted volumes that have low signal due to motion……………………………….....46
2.9 (a) The effect of motion on the normalized WBH-FA of the basic diffusion sequence
(both before and after retrospective motion correction with FLIRT and SPM) for a
particularly restless subject, (b) normalized WBH-FA of the navigated sequence with
prospective motion correction, both without and with reacquisition, for this data with
many uncorrected corrupted volumes (both before elimination (BE) of uncorrected
corrupted volumes) (c) normalized WBH-FA of the navigated sequence after elimination
(AE) of uncorrected corrupted volumes………………………………………..………..47
ix
A.1 Comparison of the normalized whole brain histogram (WBHs) for one subject for an at
rest scan (NoMo) acquired with the standard diffusion sequence (basic) compared to a
scan for the same subject with motion (Mo), with retrospective motion correction, with
and without rotating (Ro) the diffusion table…………………………………………….55
A.2 Comparison of WBH-FAs for the first healthy adult volunteer for acquisitions without
motion acquired using the basic and navigated sequences, without and with eddy current
correction. A correlation ratio cost function was used in the eddy_correct function in
FSL……………………………………………………………………………………….57
A.3 Comparison of WBH-FAs for the second healthy adult volunteer for acquisitions without
motion acquired using the basic and navigated sequences, without and with eddy current
correction. A correlation ratio cost function was used in the eddy_correct function in
FSL……………………………………………………………………………………….58
A.4 Comparison of WBH-FAs for the second healthy adult volunteer for acquisitions without
motion acquired using the basic and navigated sequences, without and with eddy current
correction. A mutual information cost function was used in the eddy_correct function in
FSL……………………………………………………………………………………….59
A.5 WBH-FAs for the second healthy adult subject for acquisitions without (NoMo) and with
motion (Mo) acquired using the basic sequence without and with different
implementations of motion and eddy current correction. „Corratio” denotes correlation
ratio cost function, “mutual” denotes mutual information cost function with either 6 or 12
degrees of freedom……………………………………………………………………….60
A.6 Diagram of the grid that was placed on the head coil during scanning of the 6 adult
subjects to help them control their amount of motion………..…….…...……………….61
3.1 (a) Comparison of the whole brain white matter FA histograms averaged over all six
subjects before applying an FA threshold and for the different acquisitions: the basic
diffusion sequence (NoMo_basic), the navigated sequence without prospective motion
correction (NoMo_vNav_NoCo), the basic diffusion sequence with motion (Mo_basic),
the basic diffusion sequence with motion and retrospective motion correction
(Mo_basic_retro_FLIRT), and the navigated sequence with prospective motion
x
correction and reacquisition (Mo_vNav_all). (b) The same as in „a‟ after applying an FA
threshold (FA > 0.2)…………………………………………………………….………..72
3.2 Comparison of the whole brain white matter MD histograms averaged over all six
subjects for the different acquisitions without FA thresholding: the basic diffusion
sequence (NoMo_basic), the navigated sequence without prospective motion correction
(NoMo_vNav_NoCo), the basic diffusion sequence with motion (Mo_basic), the basic
diffusion sequence with motion and retrospective motion correction
(Mo_basic_retro_FLIRT), and the navigated sequence with prospective motion
correction and reacquisition (Mo_vNav_all)…………………………………………….74
3.3 Comparison of histograms of the averaged FA for the six subjects over the whole volume
of the corpus callosum CC for the different acquisitions; the basic diffusion sequence
(NoMo_basic), with the navigated sequence without prospective motion correction
(NoMo_vNav_NoCo), the basic diffusion sequence with motion (Mo_basic), the basic
diffusion sequence with motion and retrospective motion correction
(Mo_basic_retro_FLIRT), the navigated sequence with prospective motion correction
and reacquisition (Mo_vNav_all). The histograms of the CC were generated by TrackVis
software…………………………………………………………………………………..76
3.4 Illustration of the fiber tracks of the whole CC for one subject and for the different
acquisitions after FA thresholding (i.e. FA > 0.2); (a) the at rest baseline scan with the
basic diffusion sequence (NoMo_basic), (b) the navigated sequence without prospective
motion correction (NoMo_vNav_NoCo), (c) the navigated sequence with prospective
motion correction and reacquisition (Mo_vNav_all), (d) the basic diffusion sequence with
motion (Mo_basic), (e) the basic diffusion sequence with motion and retrospective
motion correction (Mo_basic_retro_FLIRT), (f) the same as in „d‟ but without applying
an FA threshold. The number of tracks for each acquisition was 4259, 4028, 3948, 3330,
5676, and 3647, respectively………………………………………………….…………77
3.5 Comparison of the effect of FA thresholding on the volume of the CC, overlaid on the
T1-weighted mid-sagittal slice, for one subject and for the different acquisitions: (a) the
basic navigated sequence (NoMo_vNAv_NoCo) with no threshold, (b) the basic
navigated sequence (NoMo_vNav_NoCo) with FA>0.2, (c) the same as in „b‟ with
xi
retrospective motion correction, (d) the basic diffusion sequence with motion (Mo_basic)
and with FA>0.2, (e) the basic sequence with motion and retrospective motion correction
(Mo_basic_retro_FLIRT), and (f) the navigated sequence with prospective motion
correction, reacquisition (Mo_vNav_all) and with FA>0.2. FLIRT was also applied to
NoMo_vNav_NoCo with another cost function the mutual information, 12 degrees of
freedom and the volume of the CC with FA>0.2 was 3120 mm3………………..………78
3.6 Comparison of the effect of two different FA thresholding values on the volume of the
caudate, overlaid on the corresponding T1-weighted axial slice, for one subject and for
the different acquisitions: (a) the basic navigated sequence (NoMo_vNAv_NoCo) with
no threshold, (b) the basic navigated sequence (NoMo_vNAv_NoCo) with FA
xii
acquisitions. An at rest adult (A) scan acquired with the standard sequence
(A_NoMo_basic) is superimposed on the plots. These histograms were generated in
TrackVis software. (b) shows the histograms for the same scans that were generated in
Matlab and normalized for the total number of pixels …….......………….……………..98
4.4 (a) Normalized whole brain histograms (WBHs) of FA of a child who did not move
during the basic acquisition (C1_NoMo_basic) compared with those of a child who
moved during both the basic (C2_Mo_basic) and the navigated (C2_Mo_vNav_all)
acquisitions; (b) normalized WBHs of FA of an adult for an at rest acquisition using the
basic sequence (A_NoMo_basic), and for acquisitions with motion using the basic
sequence for paediatric (C2_Mo_basic) and adult (A_Mo_basic) subjects respectively;
and (c) effect of retrospective motion correction using FLIRT on normalized WBHs of
FA of a child who moved, both without (C2_Mo_basic_retro_BE) and with
(C2_Mo_basic_retro_AE) elimination of corrupted volumes, compared to the navigated
acquisition (C2_Mo_vNav_all) and the FA histogram of a child who did not move during
the basic scan (C1_NoMo_basic)………………………………………….……...……..99
4.5 Distribution of the mean FA histogram peak location for the basic and navigated
acquisitions. For the basic acquisition, the numbers indicate that all the children with
peak location 0.15 did not move during the scan (1), while all the children with peak
location 0.2 did move (2)…………………………………………………….…………102
xiii
List of Tables
2.1 Comparison of the mean of the WBH-FA parameters for all subjects for the different
scans………………………………………………………………………………….…..44
2.2 Comparison of the mean of the WBH-MD parameters for all subjects for the different
scans………………………………………………………………………..…………….45
A.1 Comparison of mean FA‟s in the first healthy adult volunteer for acquisitions without
motion acquired using the basic and navigated sequences, without and with eddy current
correction………………………………………………………………………………...57
A.2 Comparison of mean FA‟s in a second adult volunteer for acquisitions without motion
acquired using the basic and navigated sequences, without and with eddy current
correction and a correlation ratio cost function………………………………………….58
A.3 Comparison of mean FA‟s in a second adult volunteer for acquisitions without motion
acquired using the basic and navigated sequences, without and with eddy current
correction and a mutual information cost function………………………………………59
A.4 The range of motion (maximum – minimum) in each direction for each subject estimated
by PACE…………………………………………………………………………………62
3.1 Comparison of the FA histogram parameters for the whole brain white matter averaged
across all six subjects for the different acquisitions without FA thresholding in (a) and
after thresholding in (b) (FA > 0.2)……………………………………………………..71
3.2 Comparison of the MD (x10-3) histogram parameters for the whole brain white matter
averaged across all six subjects for the different acquisitions without FA thresholding in
(a) and after thresholding in (b) (FA > 0.2)……………………………………………..73
3.3 The whole brain white matter volume (cm3) compared for the different acquisitions
before and after FA thresholding (FA > 0.2)…………………………….………………75
3.4 Comparison of volumes for VOI‟s in gray matter (mm3) for the different acquisitions
before and after FA thresholding (FA < 0.2). Left amygdala (LA), left caudate (LC), left
xiv
hippocampus (LH), left pallidum (LP), right amygdala (RA), right caudate (RC), right
hippocampus (RH), and right pallidum (RP)…………………………………………….79
3.5 Comparison of the mean FA for different VOI‟s in gray matter averaged over all six
subjects for the different acquisitions before applying the FA threshold in (a), and after
thresholding in (b) (FA < 0.2)………………………………………….………………..81
3.6 Comparison of the mean FA histogram peak locations for different VOI‟s in gray matter
structures averaged over all six subjects for the different acquisitions before applying the
FA threshold in (a), and after thresholding in (b) (FA < 0.2)…………………….……..82
3.7 Comparison of the mean MD (x10-3) for different VOI‟s in the gray matter structures
averaged for the six subjects for different acquisitions without and with
thresholding........................................................................................................................83
4.1 Comparison of the mean FA histogram parameters of paediatric subjects for the whole
brain white matter averaged for each group ………………………………………...…100
4.2 Comparison of the mean MD histogram parameters for the whole brain white matter for
each group ………………………………………………..………………………..…...101
4.3 Comparison of the mean FA histogram parameters for the cerebral cortex for each
group................................................................................................................................102
4.4 Comparison of the mean MD histogram parameters for the cerebral cortex averaged for
each group ………………………………...…………………………………………....103
4.5 Comparison of the mean FA in different subcortical gray matter VOI‟s averaged for each
group…………………………………………………………………….…..………….104
4.6 Comparison of the mean FA histogram peak location in the different subcortical gray
matter VOI‟s averaged for each group ……………………………………………..…..105
4.7 Comparison of the mean MD (× 10-3) in different subcortical gray matter VOI‟s averaged
for each group………………………………………………………….....…………….106
5.1 Differences between the Leemans and Jones (2009) study and the present work……...117
xv
Preface
This thesis presents and evaluates the performance of a novel method to improve the efficacy of
in vivo DTI studies at the Cape Universities Brain Imaging Centre (CUBIC), South Africa. This
work originated due to the limitations of methods that are currently available for motion
correction in DTI. The main objective was to implement a technique that would be independent
of diffusion weighting, easy to apply, and affordable, specifically for use in paediatric scans.
This thesis includes three independent articles. These articles are to be found in chapters
two, three and four. Each chapter documents and evaluates various aspects of the methodology,
research, and findings. A comprehensive introduction provides the necessary background and
context to the work. This style facilitates direct access and concise evaluation of the different
methodologies and follows the logical progression of the work. However, as a complete
document, it contains necessary repetition due to the fact that each core chapter is being
presented as an independent article. For the purpose of thesis examination the contributions from
co-authors are given below.
Chapter one provides the background and rationale for the current study, as well as basic
MRI physics, diffusion MRI, and an overview of current methods for motion correction in MRI.
It also provides the background theory and principles of the pulse sequences being used in the
remainder of the thesis.
Chapter two is an article that, at the time of writing, is in press in Magnetic Resonance in
Medicine. The chapter presents the implementation of a novel echo planar imaging volumetric
navigator in DTI to correct motion and reacquire uncorrected corrupted diffusion volumes during
which excessive motion occurred. This chapter also demonstrates the effect of motion and
retrospective motion correction on the DTI-derived metrics (MD and FA maps) for the whole
brain of adult subjects.
André van der Kouwe and Ernesta Meintjes (supervisor) supervised and directed the work.
Dylan Tisdall and Aaron Hess together developed the initial proof of concept for an EPI-based
motion correction navigator. Khader M. Hasan provided assistance in analysing the DTI data.
The novel idea of how to extend and implement the EPI navigator to perform motion correction
in real time in DTI was that of the candidate, Alkathafi A. Alhamud. The EPI-based navigator
xvi
for motion correction for the scanner at CUBIC, its implementation in the DTI sequence,
algorithm development, the methods, data acquisition, validation, and processing were
performed by Alkathafi A. Alhamud. The article was drafted by Alkathafi A. Alhamud and
handed over to the co-authors for editorial and scientific commentary.
Chapter three is a manuscript that, at the time of writing, is under review by NeuroImage.
The chapter demonstrates the potential misinterpretation of FA and MD values that may occur as
a result of subject head motion. The analyses presented in this chapter uses automatic tissue
segmentation by FreeSurfer to explore changes in regional FA and MD measures in gray and
white matter, and compares the performance of retrospective motion correction to prospective
motion correction on these measures.
André van der Kouwe and Ernesta Meintjes (supervisor) supervised and directed the work.
Khader M. Hasan provided assistance in processing of the DTI data. The idea to investigate the
impact of motion on interpretation of abnormal DTI measures was that of Alkathafi A. Alhamud.
The candidate acquired and analysed all the data. The article was drafted by Alkathafi A.
Alhamud and handed over to the co-authors for editorial and scientific commentary.
Chapter four is also a prepared manuscript that will be submitted for publication. It
demonstrates the use of the navigated diffusion pulse sequence in young children (aged 5-6
years). The chapter compares the performance of the standard diffusion sequence to the
navigated diffusion sequence. The differences in the whole brain histograms (WBHs) of the FA
between paediatric and adult subjects are characterized. Automatic tissue segmentation with
FreeSurfer was used in this chapter to explore changes in regional FA and MD measures in
children in gray and white matter resulting from subject head motion.
André van der Kouwe and Ernesta Meintjes (supervisor) supervised and directed the work.
Barbara Laughton recruited the children, scheduled and arranged the paediatric scans. The idea
to investigate the impact of motion on DTI measures in young children was that of Alkathafi A.
Alhamud. The candidate analysed all the data. The article was drafted by Alkathafi A. Alhamud
and handed over to the co-authors for editorial and scientific commentary.
Chapter five is a comprehensive discussion that summarises the findings of the previous
chapters, highlights limitations, and suggests possible improvements to the current
implementation of the navigator. Chapter six presents the conclusions of the current study. It
summarizes all the salient points that have been detected throughout this study.
xvii
Acknowledgments
Ernesta Meintjies supervised the work and André van der Kouwe has been a supportive advisor
through out my work. They provided the context, expert ideas, advice, support, and numerous
resources required for the implementation of this work. I would like to thank them for their
significant contributions, as without them this would not have been possible. Khadar M. Hasan
offered several ideas and important points in the analysis of the DTI data. M. Dylan Tisdall and
Aaron A. Hess provided advices on the EPI navigator.
I would like to present my sincere gratitude to my family in specific my wife who
supported me during my PhD studies, to my mother and my deceased father.
The following organisations have provided the resources used in this work: The
University of Cape Town, Cape Universities Brain Imaging Centre (CUBIC), and the Athinoula
A. Martinos Center for Biomedical Imaging.
Funding was provided by the South African Research Chairs Initiative of the Department
of Science and Technology and National Research Foundation of South Africa, Medical
Research Council of South Africa, NIH grants R21AA017410, R21EB008547, R21MH096559,
R01HD071664, R01NS05574, P41RR014075, the Ellison Medical Foundation, and the
University of Cape Town.
1
Chapter 1
1Introduction
This thesis presents a novel method to correct, in real time, errors and artifacts caused by subject
head motion in Diffusion Tensor Imaging (DTI) using an EPI volumetric anatomical navigator.
In this chapter, MR physics, the basic principles of diffusion imaging, diffusion pulse sequences,
effects of motion in DTI, and current mechanisms for addressing these are presented. Chapter 2
describes the implementation and the validation of an EPI volumetric navigator (vNav) to correct
these errors, in real time, in DTI. Since prospective motion correction occurs at the onset of each
diffusion acquisition, volumes during which significant motion occur may be corrupted. The
mechanism for the reacquisition of these corrupted diffusion volumes is also described. Chapter
3 explores the effects of subject head motion on regional DTI measures, in particular fractional
anisotropy (FA) and mean diffusivity (MD), in gray and white matter using automatic tissue
segmentation with FreeSurfer. We explore the extent to which either retrospective motion
correction or the prospectively motion corrected diffusion sequence, both without and with
reacquisition of corrupted diffusion volumes, can recover the DTI measures in the presence of
motion. Chapter 4 presents differences due to motion in the DTI measures of different brain
regions in young (5-6 years) children, information regarding how much children in this age range
tend to move, and the success with which the prospectively motion corrected DTI sequence can
recover the DTI measures. Children were scanned using both the basic DTI sequence and the
prospectively motion corrected sequence with reacquisition. Chapter 5 is a comprehensive
discussion, summarising the main findings of the previous chapters and their implications for
DTI. The final chapter presents the conclusions.
2
1.1 Background and Motivation
Histological examination of the human brain reveals that it is comprised of more than 100
billion neurons that communicate with each other via axons for the formation of complex neural
networks. The structural mapping of such networks during healthy and diseased states is
important for understanding brain function. Previously, imaging of the white matter was, for the
most part, undertaken during post-mortem studies. The development of diffusion tensor imaging
(DTI) was a major breakthrough for studies of brain tissue connectivity (Basser et al., 1994). DTI
is a non-invasive technique that can measure white matter organization in vivo.
DTI is based on measuring differences in water diffusion in different directions in different brain
tissue. It requires scanning the brain repeatedly (i.e. multiple volumes of the brain where each
volume can consist of 72 slices with 2mm thickness) with different magnetic field gradients
applied in different directions. DTI, when compared to conventional MRI imaging techniques, is
characterized firstly by the application of very strong magnetic field gradients and secondly by a
lengthy scan time. This drawn out acquisition time renders DTI acquisition susceptible to subject
head motion especially in paediatric and non-cooperative subjects. Some of the factors that
determine the accuracy of the DTI measures are the quality of individual DTI volumes, the
alignment of the individual DTI volumes, and the accuracy of the diffusion weighting. Subject
motion can result in significant misalignment between diffusion gradient directions (scanner
coordinate system) and subject anatomy (subject coordinate system), resulting in artifacts and
errors in diffusion measures.
Several techniques have been proposed for motion correction in DTI, but all the existing
methods are fraught with limitations and drawbacks such as, for example, a substantial
lengthening of scan time (Kober et al., 2011). Some prospective methods require cameras and
markers (Aksoy et al., 2011), while others are limited by low signal to noise ratio at high
diffusion weightings (Benner et al., 2010). Retrospective motion correction is also widely used
(Rohde et al., 2004; Helton et al., 2006; Qiu et al.,2008; Leemans and Jones 2009; Rosas et al.,
2010).
The aim of this study was to develop a method for prospective motion correction in DTI that
overcomes many of the limitations of previous methods. To this end, we implemented a novel
three-dimensional echo planar imaging (3D-EPI) navigator into the DTI sequence. The current
3
method is not influenced by the strong magnetic field gradients used to acquire DTI data, as the
navigator images are acquired prior to application of the diffusion gradients. Furthermore, the
navigator only causes a minimal increase in scan time compared to other methods (Kober et al.,
2011). The navigator was developed on a Siemens 3 T Allegra (Erlangen, Germany) scanner at
Cape Universities Brain Imaging Centre (CUBIC) in South Africa, with prototype development
performed on a Siemens 3 T Tim Trio scanner located at the Athinoula Martinos Center for
Biomedical Imaging (Boston, MA, USA).
1.2 Theory
1.2.1 MRI Physics
The basic principle underlying magnetic resonance imaging (MRI) is the use of natural magnetic
properties; the spin phenomenon of the water proton (i.e. hydrogen nucleus), and its interaction
with the magnetic field. Any proton that is subjected to a magnetic field starts to precess around
the axis of that magnetic field at a frequency (ω0) given by the Larmor equation
0=B0 (1.1)
where γ is the gyromagnetic ratio of the nucleus being imaged (typically hydrogen) and B0 is the
strength of the static main magnetic field. The precession of millions of magnetic dipole
moments gives rise to a net magnetization vector, M, initially parallel to B0, where B0 also
defines the Z axis. To generate a signal, a transverse magnetic field (B1) needs to be applied
perpendicular to the main magnetic field B0. This transverse magnetic field is generated by
applying a radiofrequency (RF) pulse to a coil. If the RF pulse is applied at the Larmor
precession frequency, which is termed the resonance condition, it causes the net magnetization
M to precess around the direction of the transverse magnetization at frequency
1 B1, while
simultaneously precessing around B0 at the Larmor precession frequency. The net effect of these
combined motions is to flip the magnetization from the longitudinal (Z) direction into the
transverse (XY) plane with a spiralling type motion. The magnetization will continue precessing
around B1 for as long as the RF pulse is applied. The flip angle
is a function of the time t that
the RF pulse is applied and is given by equation (1.2).
2B1t . (1.2)
4
If the M is flipped completely into the transverse plane, it is called a 90 degree flip angle, while a
180 degree flip angle will invert the magnetization. When the RF pulse is turned off the
magnetization will revert to precessing only around B0 at the Larmor precession frequency.
This precessing M causes a change in flux in a receiving coil, which induces a current in the coil
according to Faraday‟s law. This current is the MRI signal.
The M does not continue to precess forever, but relaxes back to its equilibrium position parallel
to B0. Since it is only the transverse component of the magnetization that induces current in the
receiving coil, the MRI signal decreases over time. The signal is typically a decaying exponential
and is termed the free-induction decay (FID), since it is measured in the absence (ie. „free‟) of an
RF pulse. The equation of motion of the magnetization M is given by the Bloch equation (1.3).
1
0
2
)(
T
kMM
T
jMiMBM
dt
dM zyx
(1.3)
The three terms in the Bloch equation describe (1) precession of the magnetization around B, (2)
decay of the transverse magnetization with time constant T2, and (3) re-growth of the
longitudinal magnetization with time constant T1.
The decay of the transverse component of the magnetization Mxy is termed T2 relaxation, or
spin-spin dephasing. T2 relaxation is a consequence of the fact that the individual hydrogen
nuclei that comprise the net magnetization are all located in distinct chemical environments with
different neighbouring atoms. Each individual hydrogen nucleus will therefore experience a
slightly different local magnetic field, causing the precessional frequencies to vary. This
variation will cause the individual hydrogen nuclei to become progressively more out of phase,
and result in a decreasing Mxy. T2 contrast can be manipulated in the MR image by changing the
time to echo (TE), which defines how long after the RF pulse the signal will be acquired.
The re-growth of the longitudinal magnetization is termed T1 or spin-lattice relaxation. During
this process the nuclei will transfer the energy absorbed during excitation back to the lattice. The
efficiency of this process depends on how well the precession frequency of the magnetization
matches the frequency of motion of spins in the lattice. T1 contrast may be manipulated in the
5
images by varying the repetition time (TR), which defines the time between successive RF
pulses.
Spatial localisation or encoding is achieved in MRI by applying three mutually orthogonal field
gradients at different times during the imaging sequence. Arbitrary orientations of the linear
gradients can be achieved through the simultaneous application of fixed-orientation gradients in
three axes, X, Y, and Z. The basis of spatial localisation is the fact that the nuclear precession
frequency is directly proportional to the magnetic field (eq. 1.1). The gradients induce linearly
varying magnetic fields, resulting in spatially encoded precessional frequencies. For example, by
applying the gradient during excitation, only spins that match the resonance condition will be
excited. In this way an imaging slice is selected. Similarly, by applying the gradient during
readout, the location from which signal with a specific frequency originated can be determined,
yielding a „profile‟ image. In order to produce a topographic image, it is necessary to acquire
multiple profile images, so that signal excitation needs to be repeated many times with
allowances for relaxation time between excitation. The program that controls the timings and
order of the different components of the imaging sequence is termed a pulse sequence.
1.2.2 Pulse Sequences
An MR pulse sequence diagram is a simple way of showing how the RF and gradients are
applied during an acquisition. The vertical axis represents amplitude and the horizontal axis is
time. Enormous numbers of pulse sequences have been developed, each with its own application.
Some of the most widely used sequences for diffusion imaging are described briefly below
(McRobbie et al., 2007).
Gradient Echo (GE)
The gradient echo (GE) sequence is the most basic MRI sequence. It comprises a series
of excitation pulses that are each separated by a repetition time TR. Data is acquired at a specific
time after the application of the excitation pulse and this is defined as the echo time TE. TE is
defined as the time between the mid-point of the excitation pulse and the mid-point of the data
acquisition window. The contrast in the image will vary with changes to both TR and TE. GE is
suitable for fast imaging due to the low flip angles that can be used, facilitating ultra-short TR‟s.
6
The disadvantages of this sequence are difficulty to generate good T2 contrast, sensitivity to B0
inhomogeneities and sensitivity to changes in susceptibility. Figure 1.1 shows the basic elements
of the standard gradient echo sequence.
Figure 1.1: The basic elements of the gradient echo MR imaging sequence. Amplitude is shown
vertically, time horizontally. RF is the excitation pulse, Gss is the slice selective gradient, GPE is
the phase encoding gradient, GFE is the frequency encoding gradient. The sequence is repeated
for many different values of GPE.
Spin Echo (SE)
The SE sequence is similar to the GE sequence with the exception that there is an
additional 180o refocusing pulse present. The additional 180
o pulse is placed exactly halfway
between the excitation pulse and the echo (Fig. 1.2). The 180o pulse is usually applied in
conjunction with a slice selective gradient in the Z direction. The advantages of this sequence are
high signal-to-noise ratio (SNR), true T2 weighting, and relative insensitivity to the effects of B0
inhomogeneity. The major disadvantage is longer scan time.
7
Figure 1.2: Spin-echo sequence with the 180o refocused RF pulse. Following the RF pulse, the
spins are initially in phase (b) but dephase during the period labeled (c). At time (d), a 180o pulse
is applied. During period (e) the spins recover their phase. The signal is maximal at (f)
whereafter the spins again dephase.
Crusher Gradients
When a shaped RF pulse is applied together with a field gradient (Fig 1.1) it is called a
selective RF pulse. Selective slices are not uniform in the selection direction, but have a variation
in flip angle perpendicular to the image plane. As a result there will generally be signal from
regions with unwanted magnetization included in the FID signal.
For the 180o refocusing pulse a pair of „crusher‟ gradients can be used to eliminate any unwanted
magnetization. These two gradients are equal in size and they are placed on either side of the
refocusing pulse. The refocused magnetization that generates the echo will experience both
gradients, and because of refocusing, will remain unaffected. By contrast, the unwanted
magnetization produced by an imperfect refocusing pulse will be dephased by the single crusher
gradient lobe that it experiences. The possibility of crushers on the readout (X) and phase
encoding (Y) directions also exists. Figure 1.3 shows the spin echo sequence with the crusher
gradients placed on both sides of the 180o RF pulse (i.e. in the Z-direction only).
8
Figure 1.3: The spin-echo pulse sequence with the crusher gradients on both sides of the 180o
refocusing RF pulse. The crusher gradients are used to dephase the unwanted magnetization (FID
signal) created by the refocusing pulse.
Echo Planar Imaging (EPI)
For both the GE and SE sequences, the entire sequence needs to be repeated for a series
of phase encoding steps in order to generate the whole image. This increases the scan time and
also renders the data subject to motion artifacts. Echo planar imaging (EPI) is one of the fastest
acquisitions in MRI. Single-shot EPI fills the entire k-space matrix in a zig-zag fashion after a
single excitation. Each image can be acquired in less than 100 ms, albeit with low spatial
resolution. Since a slice is imaged in a single shot, motion is effectively frozen during the
acquisition. EPI is now the gold standard for clinical diffusion MRI (Le Bihan et al., 2006).
Depending on how the phase-encoding gradient is applied, the EPI can be either blipped or
unblipped. In blipped EPI, which is used in the current DTI sequence, a small phase-encoding
gradient „blip‟ is placed at each readout gradient reversal. This blip is of constant size and adds
further phase encoding to the previous blip making the acquisition easier and quicker. While in
the unblipped EPI, a constant phase-encode gradient is continuously used during the oscillating
read out gradient (Fig. 4b).
9
Figure 1.4: (a) In a „Blipped‟ single-shot GE-EPI sequence a small phase-encoding gradient
„blip‟ is placed at each readout gradient reversal; (b) In a „Unblipped‟ single-shot GE-EPI a
constant phase-encode is applied continuously along the readout gradient reversal. In both
modalities the whole image is acquired following a single excitation.
10
There are several drawbacks to EPI acquisition; among these are the N/2 ghost and the high
sensitivity to chemical shift artifacts; i.e. fat-shift, in the phase-encoding direction, due to the
comparatively low bandwidth in that orientation (Nagy and Weiskopf, 2008). For these reasons,
effective fat suppression and phase correction are important factors in EPI.
Fat Suppression
Hydrogen in fat has a lower Larmor frequency than hydrogen in water and this difference
is called the chemical shift. It is approximately 3.5 ppm. In order to improve fat/water soft-tissue
contrast it is often useful to „null‟ or suppress the signal from the fat.
Several methods have been developed to achieve fat suppression. Most of these rely on either
relaxation time differences between water and fat or differences in the resonant frequencies of fat
and water (chemical shift). The latter was used in the current DTI pulse sequence. This method
exploits the chemical shift between fat and water to initially excite only the fat protons. A narrow
range of RF frequencies centred on the Larmor frequency of fat is used to flip the hydrogen
nuclei in fat through 90o, leaving the water protons unexcited. This is known as the CHEmical
Shift Selective (CHESS) pulse sequence. Crusher gradients are applied immediately after the
CHESS pulse to dephase the transverse fat magnetization (Fig. 1.5). Since there is no remaining
net magnetization from fat, fat will not contribute to the signal generated by the subsequent
imaging sequence.
11
Figure 1.5: CHESS (CHEmical Shift Selective) sequence diagram. The x axis is the time in µs,
the y axis is the magnitude. From top to bottom, the crusher gradients are applied in all three
directions, X,Y and Z, respectively. The bottom trace illustrates the timing of the fat selective
(CHESS) RF pulse. The crushers are used to destroy the phase coherence of transverse fat
magnetization. This diagram was generated by the Sequence Development Environment
Software (IDEA) provided by Siemens by simulating the DTI pulse sequence.
Phase Correction
Another serious artifact that is associated with EPI is the N/2 ghost. This ghost occurs
because of cumulative phase differences between odd and even echoes over the echo train caused
by eddy currents from the rapidly switching readout gradients. Phase correction can be achieved
from a reference scan that is identical to a full scan but without the phase encoding gradient.
1.2.3 Diffusion
Diffusion is defined as the random movement of gas or liquid molecules through thermal
agitation as a function of temperature above 0 Kelvin. In pure water, collisions between
molecules cause random movement without any preferred direction, this is known as Brownian
motion (Brown R., 1828). Over time, this random “walk” process produces net displacement.
These displacements are randomly distributed when considering large molecular populations (Le
Bihan et al., 1991). The probability that a molecule travels a distance r during a time interval t
can be estimated. When the molecules move equally and freely in all directions, the diffusion is
referred to as having a Gaussian distribution with zero mean, because the probability of
12
movement in one direction is the same as movement in any other direction. The variance of the
distance travelled in all three dimensions is proportional to the time interval t, according to the
so-called Einstein equation (equation 1.4).
Dtr 62 (1.4)
D is known as the diffusion coefficient and characterizes the mobility of molecules within, and
relative to, the diffusion medium which is expressed in mm2 s
-1. Fick‟s law demonstrates that
diffusion also occurs from a region of higher concentration to a region of lower concentration. In
biological tissue, there is a high probability that water molecules interact with structures such as
cell membranes and macromolecules that reduce or impede their motion. Water exchange
between intercellular and extracellular compartments, as well as the shape of extracellular space
and tissue cellularity, affects diffusion. The term apparent diffusion coefficient (ADC) represents
the measured diffusion constants in tissues and is commonly reported in units of cm2 s
-1 or mm
2
s-1
. In the human brain, random translational diffusion of water molecules can be either isotropic
or anisotropic. Isotropic diffusion occurs when there is no preferred direction for molecular
motion such as in CSF and gray matter. In this case, the measured apparent diffusivity is largely
independent of the orientation of the tissue at the voxel length scale and it is usually sufficient to
describe the diffusion characteristics with a single (scalar) apparent diffusion coefficient (ADC).
Anisotropic diffusion occurs when water molecules diffuse more along certain directions in
comparison to others. In brain white matter fibers the presence of cellular membranes and the
surrounding myelin sheaths hinder the motion of water molecules perpendicular to the fibers. In
this case, the measured apparent axial diffusivity (parallel to tracts) is higher than radial
diffusivity (perpendicular to tracts).
1.2.4 Diffusion Weighted Imaging (DWI) Pulse Sequence
All MRI pulse sequences are, to some extent, sensitive to diffusion, but it is not possible
to produce a measurable signal attenuation that reflects the diffusion directly. To make MRI
pulse sequences sensitive to the diffusion of water molecules, additional gradient pulses are
introduced. A significant improvement in diffusion measurements was observed when a single
spin echo sequence incorporating very large but short diffusion gradients (hundreds of gauss per
centimetre during a few milliseconds) was introduced by Stejskal and Tanner (1965). The
gradient pulses are placed on each side of the 180o spin echo refocusing pulse (Fig 1.6) and are
13
balanced for "static" spins. Signal attenuation due to residual B0 field inhomogeneities become
negligible in comparison with the signal loss that occurs in moving molecules as a result of
dephasing due to the large diffusion gradients This allows accurate measurements of very small
diffusion coefficients (Le Bihan et al., 1991).
Figure 1.6: Illustration of the main components of the Stejskal and Tanner spin echo diffusion
sequence that is designed to measure the diffusion coefficient along the direction of the diffusion
gradient. δ is the duration of the diffusion gradient, Δ is the center-to-center spacing, and G is the
magnitude of the diffusion gradient in mT/m. Other sequence elements were omitted for clarity.
The effect of the diffusion gradient pulses is to induce a spatially-dependent phase shift in each
excited nucleus. The phase shift is proportional to the magnetic field. As such, for stationary
spins, the second gradient lobe will rephase the spins that were dephased by the first gradient
lobe. For spins that have moved between the application of the two gradient lobes, there will be a
net phase shift proportional to the distance travelled. This phase shift results in signal attenuation
and is given by equation (1.5).
).exp(0 ADCbSS (1.5)
where S0 is the signal intensity without diffusion weighting (termed the b0 image), S is the
attenuated signal, b is a measure of the diffusion weighting (DW), and ADC is the apparent
14
diffusion coefficient. The factor b depends on the strength, duration, and interval between the
diffusion gradients and is given by equation (1.6).
)3/(222 Gb . (1.6)
In equation (1.6), γ is the gyromagnetic ratio for hydrogen, G is the strength of the diffusion
gradient pulse in mT/m, δ is the duration of the pulse, and Δ is the center-to-center spacing. (∆-
δ/3) is known as the diffusion time t and is related to molecular motion through the Einstein
equation (1.4). To extract information about the diffusion properties of a particular voxel, the
diffusion coefficient (ADC) is calculated by rearranging equation (1.5). To obtain an image of
ADC values at least two acquisitions are necessary. Since diffusion is directional, three
orthogonal measurements are often performed and the images averaged to produce an image
with isotropic diffusion weighting (Wong et al., 1995; Mori and Zijl, 1995; Jones et al., 1999).
Due to the long diffusion encoding period, diffusion imaging is highly sensitive to motion
artefacts. Since diffusion MRI aims to calculate the diffusion of water over distances on the order
of tens of micron, any bulk motion (such as subject motion) or even involuntary physiological
motion (such as cardiac pulsation) can significantly degrade the image quality. If diffusion
weighted imaging (DWI) data is obtained using traditional multi-shot image acquisition
strategies such as GE or SE sequences, each acquisition (TR) will have a unique phase shift due
to motion between diffusion gradient lobes. This introduces a significant ghost in the
reconstructed image along the phase encoding direction (Fig. 1.7). For this reason, it becomes
necessary to use single-shot acquisition echo planar imaging (EPI) or navigator phase correction
for in vivo DW-MRI.
Figure 1.7: Typical motion artefact: ghost and signal variation across diffusion images (Le
Bihan et al., 2006).
The diffusion gradient pulse is relatively strong causing a sizeable eddy current to be generated
in the diffusion gradient coils. This eddy current produces a geometric distortion in the diffusion
http://en.wikipedia.org/wiki/Gyromagnetic_ratio
15
images (Fig. 1.8). Several diffusion-weighting preparations have been proposed in the past that
aim to reduce the effective eddy currents during data acquisition by incorporating additional
gradient lobes, often in combination with extra RF pulses. These additional gradient lobes cause
their own eddy currents, but their polarities and durations are adjusted to reduce the cumulative
eddy current effect of all gradient pulses involved. The twice-refocused spin echo (TRSE) pulse
sequence (Reese et al., 2003) successfully reduces the cumulative eddy current effect by
adjusting the timings of the gradient lobes (Fig. 1.9), at the expense of increased echo time, and
hence somewhat decreased SNR.
Figure 1.8: Images showing geometrical distortions resulting from eddy currents that are
associated with the use of strong diffusion gradients: contraction (top right), shift (bottom left),
and shear in the bottom right (Le Bihan et al., 2006).
Figure 1.9: Diagram of the standard twice-refocused spin echo (TRSE) sequence that
successfully reduces the effect of eddy currents (Reese et al., 2003). Shown are the 90o RF pulse
for excitation and the two 180o refocusing RF pulses. G is the strength of the diffusion gradients.
The duration of the diffusion gradients are δ1, δ2, δ3, and δ4, respectively, and the data is acquired
using an EPI readout. Other sequence elements were omitted for clarity.
16
1.2.5 Elements of a DWI sequence
In diffusion imaging many of the sequence elements that were discussed in preceding
sections are used in combination to overcome different obstacles that are associated either with
the diffusion gradients or with EPI. Figure 1.10 illustrates how these different elements are used
in a DWI sequence. The DWI sequence employs a CHESS pulse to perform fat suppression (a in
Fig. 1.10), an extra readout to correct for phase differences between even and odd echo‟s in the
EPI readout (b in Fig. 1.10), crusher gradients that are placed on both sides of the two 180o
refocusing RF pulses to destroy unwanted magnetization that arises from the refocusing pulses (c
in Fig. 1.10), two 180o RF refocusing pulses to overcome B0 inhomogeneity (d in Fig. 1.10), and
diffusion gradients implemented using the twice-refocused technique to overcome eddy current
effects (e in Fig. 1.10). The EPI readout with blipped phase encoding is indicated by f in figure
1.10.
Figure 1.10: Illustration of the sequence elements of the single-shot DTI pulse sequence used in
the current study. The elements are (a) the CHESS pulse, (b) additional readout for phase
correction, (c) the crusher gradients, (d) two 180o refocusing pulses, (e) diffusion gradients
implemented using the twice-refocused pulse technique. The elements of the blipped EPI readout
are indicated by f. The x axis represents the time and the y axis represents the magnitude of the
different components. This diagram was generated by the Sequence Development Environment
Software (IDEA) provided by Siemens by simulating the DTI pulse sequence.
17
1.3 Diffusion Tensor Imaging (DTI)
1.3.1 From DWI to DTI
As already mentioned, diffusion in the brain white matter is not the same along every
direction due to the presence of natural barriers. In white matter the principal barrier to diffusion
is the myelin sheaths of axons. Bundles of axons, known as fiber tracts, run in parallel creating
an environment in which water diffuses more along the length of the nerve fibers than in
directions perpendicular to them. Diffusion Tensor Imaging „DTI‟ (Basser et al., 1994), which is
an extension of diffusion MRI, exploits the anisotropic properties of diffusion in white matter to
map the white matter tracts in the brain. A single apparent diffusion coefficient can no longer be
used to characterize the anisotropic diffusion. The diffusion coefficient D or ADC is now
represented by a second rank tensor D (1.7) defined by
DDD
DDD
DDD
zzzyzx
yzyyyx
xzxyxx
D 1.7
Since the tensor is symmetric (i.e. Dyx=Dxy, Dzx=Dxz, and Dzy=Dyz), the diffusion tensor can be
calculated from a minimum of 6 diffusion-weighted volume images each with diffusion
weighting applied along a different direction.
In each voxel of the image it is important to determine the principal directions of diffusion,
called eigenvectors, and the diffusion values, called eigenvalues. When the diffusion tensor D is
fully diagonalized and the covariance between displacements in orthogonal directions is zero, the
diagonal elements of the tensor are equal to the eigenvalues λ1, λ2, and λ3. Many parameters can
be derived from the eigenvalues, in particular fractional anisotropy (FA) and mean diffusivity
(MD). FA is one of the most widely used parameters derived from DTI acquisitions (Pierpaoli
and Basser, 1996) and provides information about the microstructural integrity of highly oriented
microstructures. FA is defined in equation (1.8) and MD in equation (1.9).
18
)(2
)()()(
2
3
2
2
2
1
2
32
2
31
2
21
FA , and 1.8
3
)(321 MD 1.9
Diffusion is considered isotropic if the three eigenvalues are equal, and anisotropic, if the
eigenvalues are different in magnitude. The FA values range from one (absolutely directed
diffusion) to zero (completely isotropic diffusion). MD is a measure of average molecular motion
independent of any tissue directionality.
The main directions of the diffusivities are exploited in Tractography (Basser et al., 2000)
to reconstruct the trajectory of white matter pathways and to derive connectivity among different
parts of the brain. There are several software resources that are freely available on the internet for
fiber tracking as well as to generate all the diffusion maps, such as Diffusion Toolkit and
Trackvis (http://trackvis.org/). Trackvis, which was used in the current work, is based on
deterministic tractography. Deterministic tractography relies on the hypothesis that the main
eigenvector that corresponds to the highest eigenvalue is parallel to the underlying dominant
fiber orientation in each voxel. Based on this assumption, a single pathway along the direction of
maximum water diffusivity is propagated.
1.3.2 Effects of Subject Head Motion in DTI
Although echo planar imaging (EPI) has been implemented in the diffusion pulse
sequence mainly to minimize the effect of subject motion by acquiring the whole image within a
single shot, motion artifacts may reappear in the diffusion volume images when scanning the
brain repeatedly with different gradient directions for calculating the diffusion tensor.
Head motion causes misalignment of the diffusion volume images and furthermore, individual
voxels are exposed to a slightly different diffusion encoding direction/gradient than the desired
one (Aksoy et al., 2008). Motion during diffusion sensitization gradients also causes signal
dropouts in the images (Fig. 1.11). Both uncorrected diffusion volumes and errors in diffusion
encoding can cause serious artifacts in DTI, which can result in erroneous estimations of the
diffusion tensor information such as FA and MD.
http://trackvis.org/
19
Figure 1.11: Data affected by signal dropouts are indicated by circles. Information for these
slices is typically lost, leading to reduced SNR or biased results. The DTI acquisition for this
example consisted of 34 measurements or diffusion volumes, each with 72 slices, with a
maximum b value of 1000 sec mm-2
. The current volume was # 7.
1.3.3 Motion Correction in conventional MRI
Subject movement during MR examination renders MR data inaccurate. Consequently, a
range of techniques have been developed to correct human movement that occurs both
voluntarily (such as subject head motion) and involuntarily (such as cardiac pulsation). These
techniques are able to measure and correct the subject specific frame of reference to varying
degrees during a scan. A variety of tools or methods are available to accomplish this. One set of
methods requires modifying the pulse sequence to include a navigator. A second method utilizes
information inherent in the MRI images themselves, while a third method requires additional
hardware and software such as external optical tracking devices and microcoils. We present
below a review of these three different approaches to motion correction in MRI.
Navigator Techniques in MRI
Ehman and Felmlee (1989) first suggested that a navigator could be used with interleaved
excitation to produce a one dimensional (1D) projection of a plane. The 1D displacement along
the readout direction could then be determined by using an image space cross-correlation or k-
20
space phase roll. 1D navigators are widely used in cardiac MRI to track the motion of the right
hemi-diaphragm or the heart directly in order to correct for respiratory motion.
Subsequently, orbital navigators were developed by Fu et al. (1995). Their aim was to
sample a circle of k-space at a prescribed radius and plane. Since a rotation is equivalent in k-
space and image space, a rotation around the prescribed axis can be measured using cross-
correlation of two acquisitions. Out of plane rotation that may occur, however, invalidates the
procedures and thereby greatly reduces the efficiency of this method. By combining these two
principles, van der Kouwe et al. (2006) developed the clover navigator with 3D k-space
traversal, measuring three orthogonal arcs and three orthogonal lines through the centre of k-
space. By co-registering the clover navigator to a reference map, one can solve the six degrees of
freedom in a rigid body transformation completely. Welch et al. (2002) proposed a different
method that measures all six degrees of freedom by fully sampling a sphere of k-space. Rotations
are estimated from sphere to sphere co-registration and translations from the linear phase roll
between each navigator acquisition.
Although the previous navigators were very fast, imaging navigators with small flip
angles have been proposed for situations where there is a long preparation time before the start of
the next RF excitation pulse. Interference with the magnetization relaxation process is avoided
by using a very small flip angle of the order of 2 to 8 degrees. Spectroscopy and Magnetization
Prepared Rapid Gradient Echo (MPRAGE) sequences (Mugler and Brookeman, 1990) have been
modified in this way to generate a complete image or set of images during the magnetization
recovery/preparation time. Two previously reported imaging navigators include PROspective
MOtion (PROMO) and a 3D EPI volumetric navigator (vNav). PROMO proposed by White et
al. (2010) uses spiral imaging for the acquisition of three orthogonal images that are then co-
registered to a map of such spiral images. In order to achieve sufficient accuracy and stability
PROMO acquires a set of five navigators every 100 ms in imaging applications or five every 300
ms in Single Voxel Spectroscopy (SVS). The 3D EPI imaging navigator has been implemented
to correct for motion in SVS by Hess et al. (2011) and in a multiecho MPRAGE sequence by
Tisdall et al. (2011). This method acquires a complete three dimensional low-resolution EPI
navigator within each TR.
21
Methods that use Information Inherent in the Images
Motion correction can also be achieved using information inherent in the MRI acquisition
itself. One such method termed Periodically Rotated Overlapping ParallEL Lines with Enhanced
Reconstruction (PROPELLER) (Pipe, 1999) acquires data in k-space in strips of parallel lines,
where the lines acquired are the lowest phase-encoded lines in any Cartesian-based collection
scheme. These strips are acquired for many different directions so that the centre of k-space is
oversampled, yielding a propeller-like sampling pattern. The strips acquired in the centre portion
of k-space for each of the acquisitions are used to construct low-resolution images that can be
used to correct in-plane translation and phase errors. PROPELLER requires increased scan time
and provides limited information at the corners of k-space.
Thesen et al. (2000) developed a Prospective Acquisition CorrEction (PACE) technique for
functional MRI (fMRI). In fMRI multiple complete image volumes are acquired in rapid
succession, approximately every 1 s to 3 s. Each successive measurement can be registered to the
first volume using image volume based registration (Friston et al., 1995).
Additional Hardware for Motion correction
While most of the previous techniques require adding a „navigator‟ acquisition to the
pulse sequence or a specific strategy for data collection and acquisition, an alternative approach
is to use optical tracking devices. This can be achieved using equipment ancillary and external to
the MRI scanner namely, such as stereo cameras (Qin et al., 2009; Zaitsev et al., 2006),
retrograde reflectors with a single camera (Zaitsev et al., 2010a), or laser positioning of retro
reflectors (Eviatar et al., 1999). Cameras or lasers need to be firmly fixed inside the MR area
onto the head coil (Qin et al., 2009) or outside the scanner bore (Eviatar et al., 1999; Zaitsev et
al., 2010a; Zaitsev et al., 2006) and must be in line of sight of a specific marker or set of markers
that are affixed to the head of the subject. These systems are costly and awkward to set up due to
the presence of additional hardware and the adjuncts to the subject‟s head. Moreover the
practicality of these systems is hindered by the head coil which affects the field view available,
as well as the actual space limitations of an MR environment.
Microcoils or locator coils (Derbyshire et al., 1998; Ooi et al., 2009) are active markers
that determine the position and orientation of the head by affixing three microcoils to the
22
subject‟s head. These microcoils are tuned to receive the MR signal from a water-filled bead
within the coil.
1.3.4 Navigator Techniques in Diffusion MRI
The navigator echo technique was implemented in diffusion imaging by Ordidge et al.
(1994) to correct for subject motion even before the invention of fast imaging. They inserted a
one-dimensional navigator in the readout direction in a conventional 2DFT pulsed gradient spin-
echo (PGSE) sequence. The variations of the navigator echo phase from one view to another
were used to correct phase errors induced by translational motion in 2DFT. An improvement to
the „navigator echo‟ technique, which allowed correction not only for translation but also for
rotation of the measured object, was implemented by Anderson and Gore (1994) and De
Crespingy et al. (1995).
In these early navigator methods the „navigator echo‟ was implemented in only one
dimension (e.g. in readout direction), which caused a lack of rotational motion correction in
other dimensions. This can be overcome by acquiring a second navigator echo in the phase-
encoding direction. A 2D spiral navigator was proposed by Butts et al. (1997) in an interleaved
echo-planar imaging (IEPI) sequence. The information from the navigator, which consisted of
the constant and linear phase shifts across the head in both the x and y directions, was used to
regrid the image data